Autonomous Driving in Urban Environments: Boss and the Urban Challenge Journal of Field Robotics Special Issue: Special Issue on the 2007 DARPA Urban Challenge, Part I Volume 25, Issue 8, pages 425–466, August 2008 CMU, GM, Caterpillar, Continental, Intel Chris Urmson, Joshua Anhalt, Drew Bagnell Chris Urmson, Joshua Anhalt, Drew Bagnell, Christopher Baker, Robert Bittner,M. N. Clark, John Dolan, Dave Duggins, Tugrul Galatali, Chris Geyer, Michele Gittleman, Sam Harbaugh, Martial Hebert, Thomas M. Howard,Sascha Kolski, Alonzo Kelly, Maxim Likhachev, Matt McNaughton,Nick Miller, Kevin Peterson, Brian Pilnick,Raj Rajkumar, Paul Rybski, Bryan Salesky, Young-Woo Seo, Sanjiv Singh, Jarrod Snider,Anthony Stentz, William Whittaker, Ziv Wolkowicki, Jason Ziglar Hong Bae, Thomas Brown, Daniel Demitrish, Bakhtiar Litkouhi, Jim Nickolaou, Varsha Sadekar, Wende Zhang,Joshua Struble and Michael Taylor, Michael Darms, Dave Ferguson Presenter Fan Shen
OUTLINE • Introduction • Moving Obstacle Detection and Tracking • Curb Detection Algorithm • Intersections and Yielding • Distance Keeping and Merge Planning • Lessons learned • Conclusion 10:29 2
Urban Challenge – Launched by DARPA(Defense Advance Research Project Agency) – Develop Autonomous vehicles – Target: US military ground vehicles be unmanned by 2015 10:29 3
BOSS – Team from CMU, GM, Caterpillar, Continental, Intel – Modified from 2007 Chevrolet Tahoe to provide computer control – Equipped by drive-by-wire system – Controlled by CompactPCI with 10 2.16GHz Core2Duo CPU – Won 2007 urban challenge 10:29 4
Sensors 10:29 5
Moving Obstacle Detection and Tracking Fix shape rectangular model Point model 10:29 6
Object classification – moving or not moving • Moving flag is set when a speed is detected – Observed moving or not observed moving • Observed moving flag is set when keep moving more than a period of time 10:29 7
Predicts the motion of tracked vehicles 10:29 8
Curb detection algorithm Wavelet-based feature extraction • • 10:29 9
Wavelet-based feature extraction 10:29 10
Wavelet-based feature extraction • Collect coefficients for the current level i • Label each coefficient with label of level i-1 • Compute using these labels 1 if y[n]- >=d i • Class(y[n], i)= 0 otherwise 10:29 11
Performance of the algorithm 10:29 12
Intersections and Yielding • Intersection-Centric Precedence Estimation • Yielding 10:29 13
Intersection-Centric Precedence Estimation 10:29 14
Yielding • T required =T act +T delay +T space • L yeild polygon= V maxlane · T required + d safety • T arrival= d crash / v obstacle • T arrival> T required 10:29 15
Distance Keeping and Merge Planning • Distance Keeping • Merge Planning 10:29 16
Distance Keeping • v cmd= K gap ·(d target- d desired ) • d desired= max(v target ·l vehicle /10 , d mingap ) • a cmd= a min+ K acc v cmd·( a max- a min) 10:29 17
Merge Planning • d merge= 12m • d obst =v 0 ·d init /(v 0 -v 1 ) • X 0 -l vehicle -X 1 >=max(v 1 ·l vehicle /10, d mingap ) • X 1 -l vehicle -X 0 >=max(v 1 ·l vehicle /10, d mingap ) 10:29 18
Lessons Learned • Sensors are insufficient for urban driving • Road shape estimation maybe replaced by estimating position relative to the road • Human level driving require a rich representation • Validation and verification of the system is an unsolved problem • Driving is a social activity 10:29 19
Conclusion • A moving obstacle and static obstacle detection and tracking system • A road navigation system that combines road localization and road shape estimation where road geometry is not available • A mixed-mode planning system that is able to both efficiently navigate on roads and safely maneuver through open areas and parking lots • A behavioral engine that is capable of both following the rules of the road and violating them when necessary • A development and testing methodology that enables rapid development and testing of highly capable autonomous vehicles 10:29 20
Questions? 10:29 21
Recommend
More recommend
